I'm super bummed about this. FrontierCode 1.1 is my favorite benchmark and this has dramatically different results, I don't trust it at all!